telosieve 0.2.0-rc.4

Read-only infrastructure instruction evaluation that refuses when trusted evidence cannot agree
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
# Implementation Plan

## M0  -  executable research contract (complete)

- [x] Define the finite service state machine and authority schemas.
- [x] Register fault classes, invariants, metrics, seeds, and baseline behavior.
- [x] Implement only enough simulator and checker to run one benign and one
  poisoned-goal scenario.
- [x] Exit: deterministic replay plus a machine-readable refusal certificate.

Evidence: `scenarios/benign.json`, `scenarios/poisoned-goal.json`, and
`tests/m0.rs`. M0 supports a declared budget of zero or one fault and rejects a
configuration whose finite hypothesis count exceeds `maximum_hypotheses`.

## M1  -  hypothesis protocol

- [x] Implement bounded general hypothesis enumeration and provenance exclusion.
- [x] Add signed-history rollback and invariant-gated reconciler baselines.
- [x] Replace shared Rust/Serde planner-checker semantics with a diverse checker
  boundary and demonstrate parser/model differential tests.
- [x] Exit: every protocol requirement has behavioural evidence and all three
  baselines replay through the public harness.

## M2  -  adversarial evaluation

- [x] Run malicious/stale goals, omitted/forged observations, weakened invariants,
  partitions, equivocation, parser differential, and correlated-fault probes.
- [x] Measure hypothesis growth, compute cost, recovery latency, unsafe approvals,
  and false refusals against every baseline.
- [x] Exit: retain reproducible raw results and report all negative results.

## M3  -  productisation decision

- [x] Compare the registered safety claim with quantified availability and
  complexity costs.
- [x] Refresh novelty, name, security, soundness, and release diligence.
- [x] Exit: explicitly proceed, narrow, or archive the project based on evidence.

Historical decision: **narrow**. Productisation and public release were blocked
at M3. The private evaluation lane was later authorized in Post-M26; production
promotion and public release remain blocked. See
[PRODUCTISATION_DECISION](PRODUCTISATION_DECISION.md).

## Post-M3  -  narrowed research (historical scope)

- [x] Define multi-principal or explicitly suspectable viability semantics,
  preregistering the fault model and falsifiers.
- [x] Replace per-hypothesis process spawning with a bounded checker service or
  batch boundary without reducing semantic diversity.
- [x] Implement an authenticated, replayable history baseline rather than the
  current fixture-level proxy.
- [x] Obtain independent environment/toolchain reproduction of the registered
  experiments.

## Post-M4  -  evidence expansion

- [ ] Obtain third-party organizational reproduction and security review.
- [x] Test correlated viability-principal faults and document fault-domain
  independence requirements.
- [x] Prototype a durable rollback-resistant history anchor.
- [x] Measure the safety/availability frontier over a larger generated state
  space.

Evidence: 512 authenticated generated scenarios and per-cell results in
[GENERATED_STATE_SPACE](GENERATED_STATE_SPACE.md).

## Post-M5  -  reproduced unsafe-approval remediation

- [x] Add a viability-independent stable-key continuity invariant to both
  checker implementations.
- [x] Preserve value updates and additions while refusing implicit key deletion.
- [x] Re-run retained fixtures and the generated state space without changing
  their safety oracles.
- [x] Exit: zero unsafe approvals across retained evidence, quantified refusal
  cost, versioned certificates/checkers, and explicit residual bounds.

Evidence: [STABLE_KEY_SAFETY_KERNEL](STABLE_KEY_SAFETY_KERNEL.md). Productisation
remained blocked at that milestone by finite-model limits, false refusals,
deletion semantics, and independent-review gates.

## Post-M6  -  goal-domain availability

- [x] Authenticate multiple agreeing goal principals and declare their fault
  domains when goal faults are in scope.
- [x] Exclude complete goal domains per bounded hypothesis and require a
  surviving goal principal.
- [x] Reject authenticated disagreement and invalid domain mappings before
  planning.
- [x] Exit: unchanged 512-scenario oracle, zero unsafe approvals, zero false
  refusals, quantified hypothesis/latency costs, and explicit independence limits.

Evidence: [MULTI_PRINCIPAL_GOALS](MULTI_PRINCIPAL_GOALS.md). Third-party
organizational reproduction/security review remains unchecked and cannot be
self-certified by this repository.

## Post-M7  -  explicit authorized deletion

- [x] Add separately signed deletion evidence bound to the exact goal and
  phenotype tip.
- [x] Require agreeing deletion principals, complete fault-domain mappings, and
  a surviving authorization domain.
- [x] Enforce exact deletion sets independently in Rust and Python and expose
  per-hypothesis authorization in certificate v6.
- [x] Exit: authorized deletion applies, ordinary omission and replay/overbreadth
  refuse, existing 512-case safety/availability remains 0/0, and costs/bounds
  are retained.

Evidence: [AUTHORIZED_DELETION](AUTHORIZED_DELETION.md). Durable one-shot
consumption and third-party organizational review remained unresolved.

## Post-M8  -  durable deletion consumption

- [x] Derive an auditable identifier from the exact authenticated deletion
  envelope set and expose it in certificate v7.
- [x] Atomically commit history-anchor advancement and applied deletion
  consumption in one versioned durable state.
- [x] Fail closed on replay, corruption, legacy state, stale locks, history
  conflicts, and the bounded ledger's capacity limit.
- [x] Preserve deterministic stateless research replay and document safe
  authorization burn when later evidence persistence fails.
- [x] Exit: the anchored authorized fixture applies once, its identical replay
  fails before new evidence is emitted, and existing local validation passes.

Evidence: [DURABLE_DELETION_CONSUMPTION](DURABLE_DELETION_CONSUMPTION.md).
Third-party organizational reproduction/security review remains an external
unchecked gate.

## Post-M9  -  transactional local reference actuator

- [x] Add explicit initialization from a verified authenticated phenotype.
- [x] Require commit-time equality with observed service state and the certified
  transition precondition.
- [x] Atomically commit service state, authenticated history, and deletion
  consumption in one bounded local store.
- [x] Emit certificate-v8 actuation receipts while preserving certificate-v7
  deterministic research runs.
- [x] Exercise apply, refusal, replay, contention, corrupt/oversized state,
  post-commit evidence failure, and the real CLI lifecycle.
- [x] Exit: the local reference service changes only after an applied certified
  plan and stale retry cannot duplicate the change.

Evidence: [LOCAL_REFERENCE_ACTUATOR](LOCAL_REFERENCE_ACTUATOR.md). Real production
service integration, platform qualification, and external organizational review
remain unresolved.

## Post-M10  -  local actuator recovery qualification

- [x] Add monotonic generations and a separate pending/committed recovery
  witness around every actuator mutation.
- [x] Add bounded create-new backup, exact-latest restore, and stale/tampered
  backup rejection.
- [x] Preserve deletion history through an explicit crash-safe schema-v1 upgrade.
- [x] Resolve commit, initialization, and upgrade interruption without guessing.
- [x] Stress concurrent writers and 16 forced terminations, and retain
  50-iteration backup/restore/recovery measurements.
- [x] Exit: each tested interruption recovers to one exact committed generation
  or fails closed; stale restore cannot roll the witness back.

Evidence: [ACTUATOR_RECOVERY](ACTUATOR_RECOVERY.md). At Post-M10, qualification
was limited to the tested macOS/aarch64 single-host filesystem. Whole-disk
rollback, other platforms, external service integration, and independent review
remained open.

## Post-M11  -  Linux recovery qualification

- [x] Run recovery-state and forced-termination tests on Linux arm64 and amd64.
- [x] Use a pinned, network-disabled multi-architecture container with the
  repository mounted read-only.
- [x] Put actuator mutations on Docker-managed Linux volumes rather than tmpfs.
- [x] Retain architecture, image, filesystem, object-size, and latency evidence.
- [x] Exit: both architectures recover every tested termination to an exact
  witnessed generation; emulated timing and VM/storage limits remain explicit.

Evidence: [ACTUATOR_RECOVERY](ACTUATOR_RECOVERY.md). This removes the
macOS-only software-path gap for the tested Linux VM/volume boundary. Bare-metal
Linux, power-loss persistence, whole-disk rollback, external service integration,
and independent review remain open.

## Post-M12  -  key and identity lifecycle

- [x] Define authority-key rotation, revocation, expiry, and compromised-key
  recovery without invalidating historical certificates.
- [x] Bind lifecycle statements to authority kind, subject, sequence, and
  predecessor state.
- [x] Reject revoked or superseded keys for new transitions while retaining
  deterministic historical verification.
- [x] Exercise stale rotation, rollback, equivocation, partial availability,
  emergency revocation, and bounded state growth.

Evidence: [KEY_LIFECYCLE](KEY_LIFECYCLE.md). Lifecycle state is bounded and
authenticated but its roots and tips remain trusted scenario configuration, not
a durable organizational identity service.

## Post-M13  -  shadow-mode external adapter

- [x] Name one external target and define a read-only phenotype/goal mapping
  without mutation credentials.
- [x] Bind observations to resource versions and reject schema or concurrency
  ambiguity.
- [x] Exercise drift, partial reads, stale watches, identity mismatch, and
  bounded observation size.
- [x] Retain shadow decisions and operator-facing refusal reasons without adding
  production actuation.

Evidence: [KUBERNETES_SHADOW](KUBERNETES_SHADOW.md). This is a bounded exported
snapshot adapter, not live Kubernetes access or actuation.

## Post-M14  -  incident and recovery exercises

- [x] Define machine-readable drills for corruption, witness loss, full ledgers,
  stale locks, bad upgrades, key compromise, and lifecycle rollback.
- [x] Require explicit expected state, operator action, recovery point, and
  evidence preservation for every drill.
- [x] Execute the drills without production credentials and retain results.

Evidence: [INCIDENT_DRILLS](INCIDENT_DRILLS.md). Seven deterministic local
software drills pass; real infrastructure and organizational response remain
outside this evidence.

## Post-M15  -  property and parser robustness

- [x] Add bounded property/fuzz corpora for authority, lifecycle, shadow, and
  recovery parsing.
- [x] Retain minimized regressions for every discovered discrepancy.

Evidence: [PARSER_CORPORA](PARSER_CORPORA.md). Four deterministic corpora pass
within explicit case and byte bounds; four minimized rejection regressions are
retained and the run discovered no additional discrepancy.

## Post-M16  -  independent-assessment handoff

- [x] Assemble a self-contained assessor manifest with commit, commands,
  expected digests, claim boundaries, and unresolved release blockers.
- [x] Verify the handoff from a clean local checkout without production
  credentials or hosted CI.

Evidence: [ASSESSOR_HANDOFF](ASSESSOR_HANDOFF.md). The frozen Post-M15 source
commit passes local CI and reproduces the two operational aggregates from a
fresh local clone; this is a handoff package, not independent assessment.

## Post-M17  -  dependency provenance and advisory inventory

- [x] Generate a locked dependency and license inventory with exact package
  versions and source/checksum provenance.
- [x] Run a local advisory audit and document unresolved, unavailable, or
  accepted findings without enabling hosted CI.

Evidence: [SUPPLY_CHAIN](SUPPLY_CHAIN.md). The deterministic inventory binds 44
packages to the lockfile; the current RustSec snapshot reports no known
vulnerability or warning and no finding is accepted.

## Post-M18  -  trusted-time and recovery-root ceremony

- [x] Model trusted-time rollback/forward failure paths for lifecycle expiry and
  emergency revocation.
- [x] Define and rehearse a credential-free multi-party recovery-root ceremony
  with explicit quorum, evidence, abort, and compromise handling.

Evidence: [RECOVERY_ROOT_CEREMONY](RECOVERY_ROOT_CEREMONY.md). Enrolled
lifecycle evaluation now requires a bounded trusted-time window and the
credential-free 2-of-3 local rehearsal exercises success, exclusion, duplicate,
missing, divergent, stale, veto, and compromised-participant failure paths.
Neither mechanism establishes a production clock, participant identity, private
key custody, or organizational independence.

## Post-M19  -  protocol compatibility and migration corpus

- [x] Retain old/new scenario and certificate compatibility vectors across
  supported protocol versions.
- [x] Define fail-closed migration rules for optional-to-required trust fields
  and unknown future versions.

Evidence: [PROTOCOL_COMPATIBILITY](PROTOCOL_COMPATIBILITY.md). Two bounded
integration vectors cover legacy/current/missing-trust scenarios, certificates
v7–v9, future fields/versions, cross-version shape confusion, and oversized
certificate input. Migration validates and regenerates from authoritative input;
it never silently rewrites retained evidence.

## Post-M20  -  compatibility consumer qualification

- [x] Exercise the supported certificate boundary in an independent downstream
  reader and retain version-by-version results.
- [x] Define a deprecation window and explicit removal gate for supported
  certificate versions.

Evidence: [PROTOCOL_COMPATIBILITY](PROTOCOL_COMPATIBILITY.md). A dependency-free
Python reader agrees with Rust on three supported versions and nine refusal
classes in 13 bounded cases. Versions v7–v9 remain active; deprecation requires
explicit evidence and approval, and removal requires two later completed
milestones plus zero registered consumers and a major compatibility decision.

## Post-M21  -  certificate evidence authenticity

- [x] Define a domain-separated certificate-attestation envelope without
  invalidating retained unsigned research certificates.
- [x] Exercise signer rotation, tampering, cross-context replay, expiry, and
  bounded verification through both supported readers.

Evidence: [CERTIFICATE_ATTESTATION](CERTIFICATE_ATTESTATION.md). Detached
attestation preserves v7–v9 bytes, binds exact certificate/context/time/key
evidence, and passes seven bounded old/new and failure cases with zero
Rust/Python disagreements.

## Post-M22  -  attestation timestamp and revocation witnesses

- [x] Prototype an append-only trusted timestamp witness that prevents
  compromised signers from backdating attestations.
- [x] Define bounded signer-revocation distribution and retained historical
  verification semantics across both readers.

Evidence: [ATTESTATION_WITNESSES](ATTESTATION_WITNESSES.md). Separate signed
timestamp chains and digest-anchored revocation snapshots pass seven bounded
historical and refusal cases with zero Rust/Python disagreements. Authorities,
trusted tips, custody, and distribution remain research configuration.

## Post-M23  -  witness durability and availability

- [x] Qualify timestamp/revocation trusted-tip persistence across crash,
  backup/restore, and independent-host boundaries.
- [x] Measure bounded witness and revocation distribution availability under
  delay, loss, partition, and authority outage.

Evidence: [WITNESS_DURABILITY](WITNESS_DURABILITY.md). Atomic dual-tip state,
exact recovery/restore, and path-independent copies pass on macOS plus pinned
offline Linux arm64/amd64. Seven deterministic bounded distribution profiles
preserve three available paths and refuse loss, outage, over-budget delay, and
equivocation.

## Post-M24  -  isolated external-endpoint harness

- [x] Orchestrate separate authenticated timestamp/revocation HTTP processes
  without adding mutation authority or weakening exact-tip verification.
- [x] Exercise bounded live delay, loss, partition, outage, restart, stale,
  equivocal, unauthorized, and oversized endpoint behavior.

Evidence: [WITNESS_ENDPOINT_HARNESS](WITNESS_ENDPOINT_HARNESS.md). Three
end-to-end success paths and seven live refusal paths pass through real HTTP and
the existing cryptographic verifier within explicit request/resource bounds.

## Post-M25  -  independently operated witness reproduction

- [x] Freeze an operator handoff and fail-closed returned-evidence validator
  without manufacturing or self-certifying organizational independence.
- [ ] Integrate one independently administered timestamp/revocation endpoint
  with documented identity, custody, clock, and incident boundaries.
- [ ] Have an independent operator reproduce durability and live network fault
  results without project-controlled credentials or execution.

Preparation evidence: [WITNESS_OPERATOR_HANDOFF](WITNESS_OPERATOR_HANDOFF.md).
The handoff binds the M24 source and required outcomes; its adversarial validator
accepts one synthetic schema fixture and refuses thirteen integrity/policy faults.
No independent record has been returned, so both external gates remain open.

## Post-M26  -  private evaluation product authorization

- [x] Authorize a production-shaped private evaluation lane with a
  machine-checked scope and claim boundary.
- [x] Define candidate readiness and production-promotion gates without
  weakening the private-repository local-CI policy.
- [x] Stabilize a versioned evaluation CLI and strictly validated configuration.
- [x] Add a fail-closed, least-privilege live Kubernetes read-only collector.
- [x] Package documented install, upgrade, rollback, backup, and uninstall
  workflows for supported evaluation platforms.
- [x] Add bounded operator diagnostics and privacy-reviewed evidence export.
- [x] Qualify platform resources, interruption, recovery, and private bundle
  reproducibility before proposing an evaluation candidate.

Decision evidence:
[EVALUATION_PRODUCT_DECISION](EVALUATION_PRODUCT_DECISION.md). The authorization
enables implementation and later private candidate packaging; it is not itself
an evaluation release or production promotion.

CLI evidence: [EVALUATION_CLI](EVALUATION_CLI.md). The v1 read-only command and
configuration pass their real process/file lifecycle plus nine fail-closed
configuration and input-bound cases. The v2 live mode adds a four-read,
time/response-bounded `kubectl` collector with pre/post controller consistency,
pod ownership/readiness checks, and a namespace-scoped example RBAC grant. Its
process harness is project-controlled simulation evidence; Post-M30 later adds
a separate real local-cluster qualification.

Lifecycle evidence: [EVALUATION_LIFECYCLE](EVALUATION_LIFECYCLE.md). A bounded
single-host macOS/Linux manager atomically activates digest-bound
binary/configuration releases, verifies owner-only backups, preserves evidence
through rollback and uninstall, and passes the real five-phase lifecycle plus
ten unsafe/tamper refusals. Package-manager, service, power-loss, and
reproducible private-bundle qualification remain later gates.

Diagnostics evidence: [OPERATOR_DIAGNOSTICS](OPERATOR_DIAGNOSTICS.md). The
offline exporter verifies the managed release and owner-only evidence under
1,000-file/256-MiB bounds, emits only an explicit digest/size allowlist, and
passes a unique-secret non-disclosure check plus four fail-closed paths. Digests
remain sensitive correlators; this is not telemetry or independent privacy
assessment.

Bundle evidence: [PRIVATE_BUNDLE](PRIVATE_BUNDLE.md). The current macOS host
records bounded build time and peak child RSS, forced pre-publication
interruption leaves no bundle, recovery succeeds, and two independently built
ZIPs are byte-identical with a verified per-entry checksum manifest. This is
candidate engineering evidence, not signing custody, Linux/Kubernetes resource
qualification, power-loss proof, or independent validation.

## Post-M27  -  signed private candidate boundary

- [x] Implement domain-separated Ed25519 signing and independently configured
  verification for the exact private bundle bytes.
- [x] Bound bundle, signature, identity, key-set, and validity-window inputs and
  refuse tamper, substitution, ambiguity, unsafe key files, and stale evidence.
- [ ] Establish an operational release-signing identity and independently
  governed private-key custody.
- [ ] Freeze, sign, and independently assess the exact evaluation candidate.

Protocol evidence: [PRIVATE_BUNDLE](PRIVATE_BUNDLE.md). Tests use an ephemeral
fixture key only. Implementing the signing boundary does not establish custody,
produce a release signature, or satisfy independent assessment.

## Post-M28  -  commit-bound candidate ceremony

- [x] Bind the canonical full source commit into the deterministic bundle.
- [x] Prove commit substitution changes the bundle and malformed identities
  refuse without publication.
- [x] Define a two-custodian-ready offline signing, verification, transfer, and
  abort procedure without creating operational credentials.
- [ ] Execute the ceremony with an approved operational identity and custodians.

Evidence: [CANDIDATE_SIGNING](CANDIDATE_SIGNING.md) and
[PRIVATE_BUNDLE](PRIVATE_BUNDLE.md). Project-controlled qualification proves the
mechanism only; it does not satisfy the operational or independent gates.

## Post-M29  -  reproducible candidate build provenance

- [x] Rebuild the locked/offline release binary in two isolated target trees and
  require byte identity under explicit time, output, and resource bounds.
- [x] Retain commit, clean-tree, binary, toolchain, target, and resource evidence
  and prove altered-binary detection.
- [x] Require the candidate ceremony to match the bundle binary to a clean-tree
  provenance record for the same commit.
- [ ] Obtain an independent and cross-platform reproducible build.

Evidence: [BUILD_PROVENANCE](BUILD_PROVENANCE.md). The qualification is
same-host and project-controlled; compiler/dependency trust and hermeticity are
not established.

## Post-M30  -  real Kubernetes end-to-end qualification

- [x] Exercise the complete evaluation CLI against a disposable real Kubernetes
  API server using namespace-scoped service-account credentials.
- [x] Prove real RBAC permits only required reads and denies mutation and Secret
  access while the target remains unchanged.
- [x] Refuse authority mismatch and real API-server outage without evidence,
  under bounded time/memory and deterministic cluster cleanup.
- [ ] Qualify managed clusters, extended capacity/load, additional versions, and an
  independently operated environment.

Evidence: [KUBERNETES_REAL_CLUSTER](KUBERNETES_REAL_CLUSTER.md). This closes the
real local-cluster gap, not the managed-platform or independent-validation gates.

## Post-M31  -  OpenTofu plan evaluation integration

- [x] Evaluate a real saved OpenTofu plan through the stable read-only CLI
  without cloud credentials, external providers, or target mutation.
- [x] Bind the exact bounded plan bytes and version/resource metadata into a
  version-specific certificate extension.
- [x] Refuse destructive, sensitive, unknown, duplicate-replica, malformed,
  oversized, or authority-mismatched plan inputs before evidence persistence.
- [ ] Qualify external providers, remote state, large plans, additional OpenTofu
  versions, and an independently operated environment.

Evidence: [OPENTOFU_PLAN](OPENTOFU_PLAN.md). The real local lifecycle uses only
the built-in `terraform_data` resource and disposable local state. It establishes
plan parsing and evidence binding, not provider correctness or apply safety.

## Post-M32  -  evaluation contract and capability reconciliation

- [x] Replace the stale single-mode contract field with an explicit bounded
  schema-to-mode inventory under a no-target-mutation authority boundary.
- [x] Publish the compiled capability inventory through the CLI and require
  exact agreement with the policy contract.
- [x] Refuse uncontracted, missing, mutation-capable, mutation-shaped,
  duplicate-schema, and authority-boundary-weakened variants.

Evidence: [EVALUATION_PRODUCT_DECISION](EVALUATION_PRODUCT_DECISION.md) and
`results/evaluation-contract-validation.json`. Contract v2 reconciles three
implemented read-only modes and passes six adversarial drift refusals. It does
not authorize production actuation, credentials, publication, or promotion.

## Post-M33  -  complete three-mode private candidate input

- [x] Package all contracted Kubernetes shadow/live and OpenTofu plan
  configurations plus their operator, threat, example, and test-plan material.
- [x] Bind the supplied binary's bounded capability inventory, contract, source
  commit, and fixed contents into deterministic bundle manifest v3.
- [x] Refuse capability mismatch, malformed or oversized output, and timeout
  without publication; retain signing and independent assessment as explicit
  unsatisfied gates.

Evidence: [PRIVATE_BUNDLE](PRIVATE_BUNDLE.md). The deterministic qualification
reproduces the three-mode unsigned candidate input, validates every relationship,
survives interruption, binds source substitution, and passes four capability
refusals. No operational identity, signature, recipient, transfer, or independent
assessment is manufactured.

## Post-M34  -  executable adversarial integration coverage

- [x] Register a bounded threat-by-mode coverage contract for every supported
  evaluation integration.
- [x] Require executable anchored evidence for every applicable required cell
  and explicit reasons for non-applicable or deferred cells.
- [x] Refuse omitted or duplicated threats, mode drift, missing required
  coverage, dishonest evidence references, excessive evidence, and weakened
  authority boundaries.
- [x] Bind the registry, validator, retained result, and open gaps into the
  deterministic private candidate input.
- [ ] Resolve compromised-consistent-producer gaps across the applicable modes.
- [x] Resolve sustained-adversarial-load gaps across the applicable modes.

Evidence: [ADVERSARIAL_COVERAGE](ADVERSARIAL_COVERAGE.md) and
`results/adversarial-coverage-validation.json`. Ten threat classes contain 23
covered cells, two explicit deferred cells, and five justified non-applicable
cells; eleven adversarial registry mutations fail closed. This inventory prevents
coverage overstatement but is not independent validation or a robustness proof.

## Post-M35  -  sustained adversarial integration load

- [x] Run 16 one-byte-over-limit attacks per evaluation mode with four-way
  concurrency, per-case deadlines, and no evidence publication.
- [x] Bound total wall time, peak child RSS, diagnostic output, process count,
  input bytes, and retained-result drift.
- [x] Run eight additional real Kubernetes evaluations at four-way concurrency
  with separate evidence paths and verify the target remains unchanged.
- [x] Promote only the three sustained-load cells to covered and retain the
  compromised-consistent-producer cells as explicit gaps.

Evidence: [SUSTAINED_ADVERSARIAL_LOAD](SUSTAINED_ADVERSARIAL_LOAD.md) and
`results/sustained-adversarial-load.json`. The project-controlled bounded sample
is not a capacity forecast, managed-platform result, denial-of-service guarantee,
or independent validation.

## Post-M36  -  authenticated observation-quorum primitive

- [x] Domain-separate Ed25519 signatures over exact observation bytes, subject,
  mode, producer identity, fault domain, and bounded validity.
- [x] Require canonical bounded trust/quorum documents and at least two distinct
  authenticated producer domains.
- [x] Exercise all three modes plus substitution, forgery, domain, time,
  identity, canonical-shape, and resource refusals.
- [x] Make quorum verification mandatory in versioned evaluation schemas and
  bind the verified evidence digest into compatible certificates.
- [ ] Establish independently operated producer domains and key custody.

Evidence: [OBSERVATION_QUORUM](OBSERVATION_QUORUM.md), compiled evaluation
capabilities, the product contract v3 validator, and focused tests in
`src/observation_quorum.rs`. M37, M39, M40 and M42 apply the primitive to every
supported evaluation mode; operational independence remains open.

## Post-M37  -  corroborated Kubernetes shadow evaluation

- [x] Require a bounded canonical trust document and signed multi-domain quorum
  in the stable Kubernetes shadow evaluation schema v4.
- [x] Authenticate the exact snapshot bytes, subject, mode, signer identities,
  fault domains, and validity window before parsing or evidence persistence.
- [x] Bind the verified quorum evidence digest into certificate v9 while
  retaining readability of historical v9 certificates without the extension.
- [x] Package deterministic synthetic examples and a fixture generator, and
  prove forged quorum refusal without certificate or ledger output.
- [x] Apply corroborated collection to Kubernetes live and OpenTofu evaluation.
- [ ] Establish independently operated producer domains and operational key
  custody; bundled fixture keys are public test material only.

Evidence: [OBSERVATION_QUORUM](OBSERVATION_QUORUM.md),
[EVALUATION_CLI](EVALUATION_CLI.md), and `tests/evaluation_cli.rs`. This closes
the configured multi-domain path for shadow, with M39 and M40 extending it to
Kubernetes live and OpenTofu. It does not prove that configured domains are
genuinely independent or truthful in deployment.

## Post-M38  -  observation producer signing boundary

- [x] Expose domain-separated observation signing for exact bounded input bytes
  through a stable CLI command.
- [x] Require an absolute owner-only, single-link Ed25519 seed file and a new,
  non-colliding output path with owner-only atomic publication.
- [x] Expose public-key derivation without disclosing the private seed.
- [x] Prove two independently invoked CLI attestations interoperate with the
  quorum verifier and refuse unsafe keys, oversized input, and output collision.
- [x] Define and integrate bounded independently collecting producer processes
  for Kubernetes-live evaluation.
- [x] Integrate separately invoked bounded renderer/signing producers for
  OpenTofu evaluation; operational independence remains a later gate.

Evidence: [OBSERVATION_QUORUM](OBSERVATION_QUORUM.md) and
`tests/observation_signature_cli.rs`. This establishes producer-side signing
mechanics, not producer independence, truthful collection, or operational key
custody; M41 completes the OpenTofu process integration without promoting an
adversarial coverage cell.

## Post-M39  -  corroborated Kubernetes-live producer processes

- [x] Replace live configuration v2 with v5 requiring a bounded trust document
  and two to eight external observation-source commands.
- [x] Require every producer's canonical snapshot to exactly equal the primary
  four-read collector snapshot before multi-domain signature verification.
- [x] Bind the verified quorum digest into certificate v9 and persist no
  evidence on disagreement, forgery, malformed or oversized output, timeout,
  invalid quorum, or path collision.
- [x] Exercise two separately invoked producer collectors against a disposable
  real Kubernetes API server and under concurrent evaluation load.
- [ ] Establish separately operated clusters/control planes, credentials, hosts,
  and key custody for independent validation.
- [ ] Integrate independent OpenTofu plan producers.

Evidence: [KUBERNETES_REAL_CLUSTER](KUBERNETES_REAL_CLUSTER.md),
[OBSERVATION_QUORUM](OBSERVATION_QUORUM.md),
`tests/kubernetes_live_cli.rs`, and
`scripts/kubernetes-observation-producer.py`. M39 detects local collector and
envelope faults but retains the compromised-producer cell because every
qualified process still trusts one API server. External independence and
production custody remain open gates.

## Post-M40  -  corroborated OpenTofu plan bytes

- [x] Replace OpenTofu configuration v3 with v6 requiring a bounded trust
  document and two to eight external exact-byte producer envelopes.
- [x] Require every producer plan to match the primary plan byte-for-byte before
  multi-domain signature verification and before plan parsing.
- [x] Bind the verified quorum digest into compatible certificate v10 evidence.
- [x] Refuse forged and disagreeing producer envelopes without certificate or
  ledger persistence while retaining destructive and authority mismatch checks.
- [ ] Establish separately operated provider/state observation paths and
  operational key custody.

Evidence: [OPENTOFU_PLAN](OPENTOFU_PLAN.md),
[OBSERVATION_QUORUM](OBSERVATION_QUORUM.md), and
`scripts/run-opentofu-plan.py`. M40 authenticates exact agreement between
configured producer processes; because the qualified producers consume the same
locally generated plan, the compromised-consistent-producer cell remains
deferred.

## Post-M41  -  bounded OpenTofu producer process

- [x] Package a reference producer that independently invokes `tofu show -json`
  on an absolute regular single-link saved plan and signs the exact bytes.
- [x] Bound saved-plan, rendered-plan, attestation, stderr, and subprocess paths
  and refuse unsafe executable, plan, and key shapes.
- [x] Exercise two separately invoked producers through stable evaluation v6 and
  refuse renderer failure/timeout/malformed/oversized output, unsafe keys,
  symlinked plans, forgery, and byte disagreement without evidence publication.
- [x] Include the producer and production-shaped example arguments in the
  private evaluator bundle.
- [ ] Establish separately operated plan generation, provider/state/backend
  access, hosts, clocks, administrators, and operational signing custody.

Evidence: [OPENTOFU_PLAN](OPENTOFU_PLAN.md), [OPERATIONS](OPERATIONS.md),
`scripts/opentofu-observation-producer.py`, and
`scripts/run-opentofu-plan.py`. M41 replaces static success fixtures with real
renderer/signing processes. The local qualification still shares one saved
binary plan and host, so it does not promote the compromised-consistent-producer
coverage cell or satisfy independent validation.

## Post-M42  -  producer identity isolation kit

- [x] Add a bounded authenticated Unix relay so the evaluator can request fresh
  observations without reading producer signing keys or platform credentials.
- [x] Define one Linux systemd service user and client group per producer with
  hardened filesystem, namespace, device, task, memory, restart, and log policy.
- [x] Package relay/client programs, Kubernetes/OpenTofu relay profiles, and
  evaluation configurations that use the socket boundary.
- [x] Qualify exact output, authentication refusal, producer failure/timeout,
  output bounds, unsafe token/configuration permissions, socket cleanup/mode,
  endpoint shape, and unit drift.
- [x] Exercise the real OpenTofu two-producer evaluation lifecycle through two
  authenticated relays and the stable CLI.
- [ ] Exercise the kit under separate real Linux UIDs and independently
  administered hosts, credentials, keys, clocks, and platform truth.

Evidence: [PRODUCER_ISOLATION](PRODUCER_ISOLATION.md), [OPERATIONS](OPERATIONS.md),
`scripts/observation-source-{relay,client}.py`,
`deploy/systemd/telosieve-observation@.service`, and
`scripts/run-producer-isolation-qualification.py`. M42 establishes a usable
deployment boundary but the local harness uses one UID and one host, so no
independence or compromised-consistent-producer claim is promoted.

## Post-M43  -  Linux producer identity qualification

- [x] Run two producer relays under distinct non-root Linux kernel UIDs and
  distinct client groups in a pinned offline container.
- [x] Give the evaluator access through only the two client groups and verify
  exact observations from both relays.
- [x] Refuse an unrelated UID, a wrong token, peer token/configuration reads,
  and peer relay access while verifying socket ownership, mode, and cleanup.
- [x] Bound the container to 256 MiB memory, 64 PIDs, 40 MiB total temporary
  filesystems, a read-only repository, and no network.
- [ ] Exercise systemd and Kubernetes through the relay on separately
  administered hosts with real credentials, clocks, keys, and platform truth.

Evidence: [PRODUCER_ISOLATION](PRODUCER_ISOLATION.md),
`scripts/qualify-linux-producer-isolation.sh`, and
`scripts/run-linux-producer-isolation.py`. This establishes real kernel-enforced
UID/GID separation on Linux. The container and host remain project-controlled,
so organizational independence and compromised-consistent-producer coverage
remain unclaimed.

## Post-M44  -  real Kubernetes relay lifecycle

- [x] Route both real-cluster Kubernetes observation producers through separate
  authenticated Unix relays and expose only relay clients to the evaluator.
- [x] Exercise the stable live CLI and eight four-way concurrent evaluations
  through the relays while preserving exact evidence and target identity.
- [x] Retain least-privilege RBAC and fail-closed authority/API outage checks,
  and add relay-outage refusal without certificate or ledger persistence.
- [x] Verify bounded relay readiness, serialization/backlog behavior, process
  termination, socket cleanup, overall time, per-case time, and child memory.
- [ ] Repeat on managed and separately administered clusters with independent
  credentials, hosts, keys, clocks, and platform truth.

Evidence: [KUBERNETES_REAL_CLUSTER](KUBERNETES_REAL_CLUSTER.md),
[PRODUCER_ISOLATION](PRODUCER_ISOLATION.md), and
`scripts/run-kubernetes-real-cluster.py`. This closes the local Kubernetes
relay-path mismatch. It does not establish managed-platform behavior,
organizational independence, or truthful observations under a compromised
shared control plane.

## Post-M45  -  candidate pre-freeze readiness evidence

- [x] Map all eight private candidate-readiness gates to exact repository files
  and authoritative local-CI commands in a strict machine-readable manifest.
- [x] Require seven implementation gates to be locally verified while keeping
  exact-byte bundle signing pending until candidate freeze.
- [x] Refuse missing, duplicate, reordered, prematurely signed,
  absent-evidence, substituted-evidence, and weakened-authority manifests.
- [x] Package the readiness manifest, validator, retained result, and claim
  boundary in the private evaluator bundle.
- [x] Freeze and sign the reviewed exact candidate bytes and emit the final
  evaluator handoff manifest.

Evidence: [CANDIDATE_READINESS](CANDIDATE_READINESS.md),
`evaluation/candidate-readiness.json`, and
`scripts/validate-candidate-readiness.py`. This closes narrative-only readiness
drift; it does not itself create or sign a release candidate.

## Post-M46  -  versioned private release-candidate freeze

- [x] Set the private evaluation candidate version to `0.2.0-rc.1` and provide
  claim-bounded evaluator-facing release notes.
- [x] Build the exact clean-master release binary and deterministic source-bound
  bundle only when its digest matches clean-tree reproducibility evidence.
- [x] Sign exact bundle bytes with an external owner-only key and emit bounded
  public trust, artifact inventory, checksums, and candidate status metadata.
- [x] Publish the seven-file handoff atomically without clobbering and provide a bound
  offline verifier for checksums, embedded commit/binary, version, and signature.
- [x] Execute the freeze at the final reviewed master commit and publish the
  private annotated release-candidate tag.

Evidence: [CANDIDATE_SIGNING](CANDIDATE_SIGNING.md),
`RELEASE_NOTES_v0.2.0-rc.1.md`, and
`scripts/{build,verify}-release-candidate.py`. Project-controlled signing
authenticates evaluator bytes but does not establish independent custody,
assessment, public release, or production promotion.

## Post-M47  -  external assessor journey

- [x] Make one current guide the canonical entrypoint for assessment of the
  exact signed `v0.2.0-rc.1` handoff.
- [x] Require independent authentication of candidate identity and public trust
  before using the source-built verifier or handling candidate contents.
- [x] Separate handoff verification, full local-CI reproduction, and the three
  environment-dependent evaluation modes with explicit prerequisites and claim
  bounds.
- [x] Distinguish direct execution of the signed candidate binary from
  source-built Kubernetes and OpenTofu harness evidence.
- [x] Define a findings record that captures identity, environment, commands,
  missing coverage, severity, reproduction, mutation observations, and
  reassessment.
- [x] Mark the Post-M15 assessor handoff as historical and route README users to
  the release-candidate procedure.
- [x] Report bounded witness-endpoint startup failures with the child-process
  diagnostic instead of an opaque readiness parse failure.
- [ ] Obtain an assessment record from an independent operator for the exact
  signed candidate bytes.

Evidence: [EXTERNAL_ASSESSMENT](EXTERNAL_ASSESSMENT.md),
[README](../README.md), and [ASSESSOR_HANDOFF](ASSESSOR_HANDOFF.md). This closes
the repository-navigation and safe-verification guidance gap; it does not create
independent evidence or authorize production promotion.

## Post-M48  -  standardized read-only integration contract

- [x] Define a versioned platform-neutral request, response, capability and
  stable refusal taxonomy for external read-only adapters.
- [x] Execute adapters through bounded stdin/stdout subprocesses without loading
  integration code into the evaluator.
- [x] Add strict evaluation configuration v7 and compiled
  `external-read-only` capability under the existing no-mutation boundary.
- [x] Require exact desired/observed authority equality and a separately signed
  multi-domain quorum over the canonical adapter response.
- [x] Bind integration, target, response and quorum identities into compatible
  certificate v11 evidence and both Rust and Python readers.
- [x] Provide a packaged example, reference conformance adapter and real-CLI
  qualification covering success, mutation declaration, partial/context/
  authority faults, timeout, output collision and repeated hostile cases.
- [ ] Qualify each future concrete platform adapter against a disposable real
  system with independently controlled read-only credentials and producers.

Evidence: [INTEGRATION_CONTRACT](INTEGRATION_CONTRACT.md),
`src/integration.rs`, `tests/integration_contract.rs`, and
`evaluation/config.integration.example.json`. This standardizes integration
mechanics; it does not qualify an unnamed platform, prove producer truth or
independence, or authorize actuation.

## Post-M49  -  concrete Redis read-only integration

- [x] Map one bounded Redis namespace into Integration Contract v1 without
  adding Redis client dependencies or mutation commands.
- [x] Load credentials only from owner-only, regular, single-link files and
  restrict this adapter version to literal loopback endpoints.
- [x] Qualify a pinned disposable real Redis 8.8 server with distinct read-only
  ACL users for the primary adapter and two signed observation producers.
- [x] Prove `SET` is denied for every evaluation credential and retain
  `target_mutated: false` through success and bounded concurrent load.
- [x] Fail closed without evidence on RESP/resource excess, 65 replicas,
  producer disagreement, stale producer timing, credential faults and complete
  server outage.
- [x] Package the adapter, producer, parser tests, real-system harness, retained
  result and operator documentation in future private evaluation bundles.
- [ ] Repeat with independently administered Redis control planes, credential
  custody and producer signing domains.

Evidence: [REDIS_INTEGRATION](REDIS_INTEGRATION.md),
`scripts/redis-integration-adapter.py`,
`scripts/redis-observation-producer.py`,
`scripts/test-redis-integration.py`, and
`results/redis-integration-qualification.json`. This qualifies a concrete local
Redis integration and real ACL enforcement; one project-controlled server does
not establish independent observation truth or production readiness.

## Post-M50  -  concrete PostgreSQL read-only integration

- [x] Map four fixed schema tables into Integration Contract v1 through a
  dependency-free bounded `psql` subprocess.
- [x] Collect one coherent `REPEATABLE READ READ ONLY` transaction with row,
  byte, statement, lock and process bounds.
- [x] Keep credentials out of arguments/environment values through owner-only
  source files and an ephemeral owner-only pgpass file.
- [x] Qualify three SELECT-only roles and prove six mutation/out-of-scope SQL
  operations are denied against pinned PostgreSQL 18.4.
- [x] Exercise a primary adapter, two signed producers, bounded concurrent load,
  committed-writer races and seven no-evidence platform faults.
- [x] Package implementation, tests, real-system harness, retained result and
  assessor/operator guidance without displacing Redis coverage.
- [ ] Repeat across independently administered PostgreSQL control planes,
  credential custody, producer keys and production topology/failover paths.

Evidence: [POSTGRESQL_INTEGRATION](POSTGRESQL_INTEGRATION.md),
`scripts/postgresql_integration_common.py`,
`scripts/run-postgresql-integration.py`, and
`results/postgresql-integration-qualification.json`. This qualifies local
transaction and privilege mechanics, not independent database truth or
production readiness.

## Post-M51  -  concrete HTTP/JSON read-only integration

- [x] Map one strict versioned HTTP JSON snapshot into Integration Contract v1
  through a dependency-free fixed-GET adapter.
- [x] Bound loopback endpoint, path, authentication, socket time, response size,
  content type, JSON shape, maps, values and replicas.
- [x] Keep bearer tokens in owner-only single-link files and out of arguments,
  diagnostics and evidence.
- [x] Qualify three bearer identities, two signed producers, eight concurrent
  evaluations and explicit denial of POST, PUT, PATCH and DELETE.
- [x] Prove no-evidence refusal for redirect, malformed, oversized, incomplete,
  timeout, authentication, producer-disagreement and outage faults.
- [x] Package the implementation, tests, harness, retained result and assessor
  guidance without displacing Redis or PostgreSQL coverage.
- [ ] Qualify audited TLS/mTLS transport, external independently administered
  endpoints, credential custody, rotation, rate limits and production topology.

Evidence: [HTTP_JSON_INTEGRATION](HTTP_JSON_INTEGRATION.md),
`scripts/http_json_integration_common.py`,
`scripts/run-http-json-integration.py`, and
`results/http-json-integration-qualification.json`. This qualifies local HTTP
protocol mechanics and failure bounds, not public-network security, external
truth, operational independence or production readiness.

## Post-M52  -  HTTP/JSON mutual-TLS transport qualification

- [x] Add explicit HTTPS without weakening loopback HTTP compatibility or
  permitting opportunistic plaintext downgrade.
- [x] Require TLS 1.3, pinned CA trust, hostname/IP verification, mutual client
  authentication and distinct adapter/producer identities.
- [x] Keep client keys owner-only, regular and single-link and keep key/token
  bytes out of arguments, diagnostics and evidence.
- [x] Qualify GET-only success and mutation-method denial through the public CLI
  and signed producer processes over an ephemeral real TLS endpoint.
- [x] Fail without evidence on wrong CA, missing/untrusted client identity,
  plaintext downgrade, disagreement, timeout and outage.
- [ ] Establish operational PKI custody, revocation checking, certificate
  rotation, external endpoints and independent assessment.

Evidence: [HTTP_JSON_INTEGRATION](HTTP_JSON_INTEGRATION.md) and
`results/http-json-integration-qualification.json`. This establishes local TLS
mechanics and downgrade refusal, not operational PKI or independent transport.

## Post-M53  -  HTTP/JSON certificate lifecycle and revocation

- [x] Version mutual-TLS credentials to require an explicit bounded CRL.
- [x] Enable leaf revocation checks in both client and server TLS contexts.
- [x] Prove a newly issued client identity can replace the original identity
  through the unchanged public evaluation CLI and signed-producer boundary.
- [x] Revoke the active server certificate and prove all readers refuse without
  certificate or ledger evidence.
- [x] Refuse missing and malformed CRL material before evidence persistence.
- [ ] Establish independently operated issuing authorities, protected custody,
  online revocation distribution, expiry monitoring and production rotation.

Evidence: `scripts/run-http-json-integration.py` and
`results/http-json-integration-qualification.json`. The qualification exercises
real OpenSSL CA state and CRLs locally; it is not operational PKI evidence.

## Post-M54  -  offline HTTP/mTLS PKI readiness

- [x] Add a bounded offline readiness command for CA, CRL, client certificate
  and private-key material.
- [x] Verify trust, CRL signature/freshness, client purpose/revocation,
  certificate renewal horizon and exact certificate/key pairing.
- [x] Emit only aggregate readiness fields without paths or secret material.
- [x] Qualify three valid identities and refuse near-expiry, mismatched-key and
  revoked-client cases.
- [ ] Connect readiness output to independently operated PKI monitoring.

Evidence: `scripts/http-json-pki-check.py` and the v4 retained HTTP result.

## Post-M55 - PKI readiness monitoring publication

- [x] Aggregate one to four offline PKI checks with stable ready/not-ready exit
  semantics and no propagated credential diagnostics.
- [x] Atomically publish an owner-only versioned status file with bounded fields.
- [x] Qualify ready to not-ready transition after certificate revocation and
  prove status output contains no credential paths.
- [x] Package a hardened systemd oneshot service and persistent timer.
- [ ] Connect the scheduler-neutral status to independently operated alerting.

Evidence: `scripts/http-json-pki-monitor.py`, the systemd units, and the v5 HTTP
qualification result.

## Post-M56 - Prometheus PKI alerting boundary

- [x] Validate exact owner-only monitor status with bounded age, future skew,
  identity counts and internal readiness consistency.
- [x] Publish six fixed-cardinality, owner-writable Prometheus gauges atomically
  without labels, paths, certificate material or checker diagnostics.
- [x] Fail closed for not-ready, stale, future, negative-time, malformed,
  inconsistent, unsafe and missing status while replacing stale green metrics.
- [x] Chain publication after every systemd monitor attempt and package the
  implementation, qualification, retained evidence and operator guidance.
- [ ] Obtain independently operated Prometheus scraping, alert routing,
  retention, on-call response and evidence.

Evidence: `scripts/http-json-pki-prometheus.py`,
`scripts/run-http-json-pki-prometheus-qualification.py`, and
`results/http-json-pki-prometheus-qualification.json`.

## Post-M59 - rc.2 release-state reconciliation

- [x] Advance the current evaluation candidate target to `0.2.0-rc.2` while
  preserving `v0.2.0-rc.1` as immutable historical evidence.
- [x] Derive active packaging and qualification metadata from the Cargo package
  version and reject version, lockfile, release-note, and assessor-guide drift.
- [x] Require the offline verifier to receive the expected candidate version
  through an independently authenticated assessor input.
- [x] Reconcile README, release strategy, assessor guidance, bundle contents,
  and curated release notes with mandatory quorum and current integrations.
- [x] Keep exact-byte status pre-freeze until the reviewed rc.2 commit is built,
  signed, verified, and tagged through a separate ceremony.

Evidence: `scripts/release_metadata.py`,
`scripts/validate-release-metadata.py`, `RELEASE_NOTES_v0.2.0-rc.2.md`, and
[EXTERNAL_ASSESSMENT](EXTERNAL_ASSESSMENT.md). This milestone makes current
release surfaces coherent; it does not sign, distribute, or independently
assess rc.2.

## Post-M60 - open-source governance and history hygiene

- [x] Add the complete Apache License 2.0 text matching package metadata and
  keep registry publication disabled.
- [x] Add security reporting, contribution, conduct, and change-history policies
  that preserve the evaluation and private-local-CI boundaries.
- [x] Record an explicit source-opening decision with separate visibility,
  hosted-CI, registry, telemetry, production, and claim gates.
- [x] Refresh exact-name, registry, domain, company, trademark-risk, prior-art,
  dependency, and advisory diligence without presenting search absence as
  legal clearance or novelty proof.
- [x] Add a bounded complete-Git-history and current-tree audit for risky paths,
  common credential shapes, owner-specific paths, and oversized artifacts.
- [x] Enforce governance files, Cargo package metadata, history hygiene, and the
  public-opening boundary in authoritative local CI.
- [ ] Change repository visibility only after productisation, an exact successor freeze,
  a final audit, and verification of the remote public state.

Evidence: [PUBLIC_OPENING_DECISION](PUBLIC_OPENING_DECISION.md),
[DILIGENCE_REFRESH_2026-08-08](DILIGENCE_REFRESH_2026-08-08.md), root governance
files, `scripts/audit-public-history.py`, and
`scripts/validate-open-source-readiness.py`. This milestone prepares public
source governance; it does not itself open the repository or activate hosted CI.

## Post-M61 - enduring product brand system

- [x] Record owner approval for one enduring product identity across research,
  evaluation, release-candidate, and any future production maturity without
  weakening production, assessment, legal, CI, or publication gates.
- [x] Freeze audience, category, purpose, promise, principles, personality,
  terminology, messaging, reasons to believe, and prohibited claims.
- [x] Retain three distinct visual directions and select the convergence gate
  through a documented non-leading internal review and small-size test.
- [x] Produce horizontal, stacked, symbol, wordmark, small, monochrome, and
  reversed SVGs plus repository, release, diagram, and chart templates.
- [x] Publish light, dark, semantic, forced-colour, typography, spacing, radius,
  icon, diagram, chart, imagery, motion, terminal, print, and channel rules.
- [x] Generate deterministic favicon, avatar, and social-card exports with a
  digest-, dimension-, colour-space-, licence-, provenance-, and command-bound manifest.
- [x] Enforce SVG safety and accessible descriptions, WCAG contrast, non-colour
  state semantics, deterministic same-host exports, README consistency,
  package inclusion, and prohibited public-copy rules in local CI.
- [x] Remove maturity from the brand version, canonical lockups, product card,
  category, and durable descriptions; retain evaluation status only in a
  separately packaged overlay and release-specific copy.
- [x] Enforce maturity-neutral canonical assets and exact permanent brand version
  in local CI, with the rc.2 label retained only as historical tag evidence.
- [ ] Obtain formal trademark/cultural review, multi-participant comprehension,
  print proofing, and cross-platform optical comparison before claiming those
  external brand assurances.

Evidence: [BRAND_IDENTITY](BRAND_IDENTITY.md),
[EVALUATION_PRODUCTISATION_DECISION](EVALUATION_PRODUCTISATION_DECISION.md),
`assets/brand/`, `scripts/build-brand-assets.py`, and
`scripts/validate-brand.py`. This completes the enduring product identity; it
does not constitute safety, independent validation, market-fit, or legal evidence.

## Post-M62 - v0.2.0-rc.2 evaluation freeze

- [x] Authorize the exact version, tag, authority boundary, signer, key ID,
  project-controlled custody label, maximum validity, recipient class, private
  channel, independent trust authentication, and non-authorized release modes.
- [x] Validate a curated product/version/theme release title, bounded highlights,
  required claim/install/evidence sections, physical-line layout, local links,
  prohibited text, and responsive HTML preview.
- [x] Package the release decision and presentation validator with the candidate.
- [x] Inspect the rendered release presentation at desktop and narrow viewports
  and retain the bounded result and residual limits.
- [x] Merge the final reviewed pre-freeze commit and pass `./scripts/ci-local.sh`
  on its clean, pushed `master` SHA.
- [x] Reproduce identical locked/offline release binaries and retain the exact digest.
- [x] Generate an external owner-only evaluation key and atomically build, sign,
  checksum, and offline-verify the seven-file handoff outside Git.
- [x] Create and push annotated `v0.2.0-rc.2` at the exact candidate commit with
  bundle/trust digests and the independent-assessment boundary.
- [x] Record and review the completed ceremony without changing candidate bytes.

Evidence before freeze: [RC2_EVALUATION_RELEASE_DECISION](RC2_EVALUATION_RELEASE_DECISION.md),
[RC2_RELEASE_PRESENTATION_REVIEW](RC2_RELEASE_PRESENTATION_REVIEW.md),
`RELEASE_NOTES_v0.2.0-rc.2.md`, and
`scripts/validate-release-presentation.py`. Exact artifacts and tag evidence
are retained in [RC2_FREEZE_RECORD](RC2_FREEZE_RECORD.md). The record commit is
deliberately newer than the immutable tagged candidate bytes.

## Post-M63 - v0.2.0-rc.3 privacy-safe successor freeze

- [x] Replace the private GitHub repository with a sanitized graph containing
  noreply-only retained identities, one branch, and no superseded candidate tags.
- [x] Rebind historical evidence to rewritten commits and pass the bounded object,
  metadata, blob, literal-address, and fresh-clone audits.
- [x] Advance package, assessor, brand-overlay, bundle, release-note, and release
  decision surfaces to `v0.2.0-rc.3` without changing the enduring product identity.
- [x] Render and inspect the release presentation at desktop and 390 px viewports,
  fix narrow-title overflow, and retain the bounded result and residual limits.
- [x] Merge the final reviewed pre-freeze commit and pass `./scripts/ci-local.sh`
  on its clean, pushed `master` SHA.
- [x] Reproduce identical locked/offline release binaries and retain the exact digest.
- [x] Generate a new external owner-only evaluation key and atomically build,
  sign, checksum, and offline-verify the seven-file handoff outside Git.
- [x] Create and push annotated `v0.2.0-rc.3` at the exact candidate commit with
  bundle/trust digests and the independent-assessment boundary.
- [x] Record and review the completed ceremony without changing candidate bytes.

Evidence before freeze: [RC3_EVALUATION_RELEASE_DECISION](RC3_EVALUATION_RELEASE_DECISION.md),
[RC3_RELEASE_PRESENTATION_REVIEW](RC3_RELEASE_PRESENTATION_REVIEW.md),
`RELEASE_NOTES_v0.2.0-rc.3.md`, [History Privacy Migration](HISTORY_PRIVACY_MIGRATION.md),
and `scripts/validate-release-presentation.py`.
Exact artifact and tag evidence is retained in [RC3_FREEZE_RECORD](RC3_FREEZE_RECORD.md).
The record commit is deliberately newer than the immutable tagged candidate bytes.

## Post-M64 - distinctive cut-sieve brand migration

- [x] Treat the owner's rejection of the faceted convergence gate as a material
  distinctiveness finding rather than a cosmetic adjustment.
- [x] Develop and compare three new directions against the mature product,
  including colour-independent silhouette, 16/24/32-pixel recognition, claim
  safety, category fit, and unintended symbolism.
- [x] Select the cut-sieve glyph and implement canonical symbol, small, favicon,
  horizontal, stacked, monochrome, reversed, avatar, social-card, and release-card
  variants without changing the wordmark, palette, tagline, or maturity boundary.
- [x] Advance the governed identity to brand `3.0.0`, archive the released `2.0.0`
  manifest, regenerate deterministic exports, and update package validation,
  README, productisation, release, and brand governance surfaces.
- [x] Keep signed `v0.2.0-rc.3` bytes and their archived brand identity immutable;
  assign brand `3.0.0` to current source and a future candidate.
- [ ] Obtain formal trademark/cultural review, multi-participant comprehension,
  print proofing, and cross-platform optical comparison before claiming those
  external brand assurances.

Evidence: [BRAND_IDENTITY](BRAND_IDENTITY.md), `assets/brand/archive/2.0.0/`,
`assets/brand/source/`, `assets/brand/concepts/`,
`scripts/build-brand-assets.py`, and `scripts/validate-brand.py`.

## Post-M65 - Fractured Oracle brand migration

- [x] Treat the owner's rejection of the cut-sieve mark as a material artistic
  distinctiveness finding and compare three non-conservative directions at equal scale.
- [x] Select Fractured Oracle, remove its literal question-mark treatment, reject
  an uncontrolled jagged fault, and approve an exact circular aperture.
- [x] Implement canonical symbol, small, favicon, horizontal, stacked,
  monochrome, reversed, avatar, social-card, and release-card variants without
  changing the wordmark, palette, tagline, or maturity boundary.
- [x] Advance the governed identity to brand `4.0.0`, retain the superseded
  `3.0.0` manifest digest, regenerate deterministic exports, and update package,
  README, productisation, release, changelog, and brand governance surfaces.
- [x] Keep signed `v0.2.0-rc.3` bytes and their archived brand identity immutable;
  assign brand `4.0.0` to current source and a future candidate.
- [ ] Obtain formal trademark/cultural review, multi-participant comprehension,
  print proofing, and cross-platform optical comparison before claiming those
  external brand assurances.

Evidence: [BRAND_IDENTITY](BRAND_IDENTITY.md), `assets/brand/archive/3.0.0/`,
`assets/brand/source/`, `assets/brand/concepts/`,
`scripts/build-brand-assets.py`, and `scripts/validate-brand.py`.

## Post-M66 - public comprehension and website source

- [x] Rewrite the README opening so a first-time reader can identify the problem,
  mechanism, concrete infrastructure example, outcome, and current limitation
  before encountering protocol terminology.
- [x] Replace the ASCII architecture sketch with one accessible, responsive SVG
  showing separate authorities, provenance verification, fault hypotheses,
  independent checking, agreement, certificate/refusal, and retained evidence.
- [x] Create a responsive dependency-free website using the permanent brand and
  current maturity overlay, with no telemetry, remote runtime assets, scripts,
  credentials, or stronger claims.
- [x] Add bounded deterministic local website assembly, digest manifest,
  internal-link and fragment checks, SVG safety checks, responsive/reduced-motion
  checks, asset budgets, and authoritative local-CI coverage.
- [x] Document that live GitHub Pages activation remains gated because even
  branch-based publishing uses a hosted deployment workflow and private-source
  Pages may be publicly accessible.
- [ ] Activate and verify the live Pages URL only after an authorizing pull
  request supplies explicit hosted-workflow approval and the required release,
  permissions, supply-chain, privacy, and public-content review.

Evidence: [GitHub Pages Website](GITHUB_PAGES.md), `site/`,
`scripts/build-pages-site.py`, and `scripts/validate-pages-site.py`.

## Post-M67 - public evaluation launch

- [x] Authorize public Apache-2.0 source, portable hosted CI, GitHub Pages,
  crates.io publication, and one prerelease GitHub Release without weakening the
  evaluation or no-target-mutation boundary.
- [x] Adopt Keep a Changelog 1.1.0 structure, advance current source to
  `0.2.0-rc.4`, enable crates.io packaging, and add curated release notes.
- [x] Add commit-pinned, least-privilege CI, Pages, and tagged-release workflows
  with bounded runtimes, no pull-request secrets, and registry credentials
  confined to the release environment.
- [x] Add public README badges, plain-language repository metadata, topic tags,
  workflow validation, and updated release and Pages procedures.
- [ ] Make the repository public and verify visibility, About metadata, topics,
  default branch, licence, remote head, and complete public history.
- [ ] Verify hosted CI and Pages deployment from the merged reviewed head.
- [ ] Publish and verify `telosieve 0.2.0-rc.4` on crates.io and docs.rs plus the
  matching prerelease GitHub Release and checksummed Linux binary.
- [ ] Inspect the live Pages site at desktop and mobile widths, verify links and
  metadata, and fix every material visual or accessibility finding.

Evidence: [Public Opening Decision](PUBLIC_OPENING_DECISION.md),
[Release Strategy](RELEASE.md), [GitHub Pages Website](GITHUB_PAGES.md),
`.github/workflows/`, `scripts/validate-hosted-workflows.py`, `CHANGELOG.md`, and
`RELEASE_NOTES_v0.2.0-rc.4.md`.